Abstract
Speech sounds are based on categories, but listeners are also sensitive to within-category acoustic variation. This gradient system helps promote flexibility, deal with ambiguity, and maintain plasticity. Although the exact sources of gradiency are not well understood, prior work suggests that it may be a product of statistical learning: Listeners exposed to more variable input show shallower categorization slopes. However, the tasks used in earlier work are ambiguous regarding what this slope represents. We investigated the role of phonetic variability as a causal driver of gradiency across three experiments, using a distributional learning paradigm in conjunction with the Visual Analog Scale task, which resolves this ambiguity. In 2024, participants were trained on distributions of voice onset time with either high or low variance, and gradiency was assessed using the Visual Analog Scale task. Experiment 1 (n = 84) suggested that variance increased trial-by-trial inconsistency, not gradiency. Experiment 2 (n = 168) used 28 items to better generalize across stimulus characteristics and found robust evidence for increased gradiency due to variability. Experiment 3 (n = 85) introduced a baseline (no learning) condition and ruled out the alternative explanation that listeners were adapting to low variance by becoming less gradient. Together, these support the idea that variable input promotes gradiency. (PsycInfo Database Record (c) 2026 APA, all rights reserved).